Hi Jason, > On Jan 14, 2015, at 20:12 , Jason Gunthorpe <jgunthorpe@xxxxxxxxxxxxxxxxxxxx> wrote: > > On Wed, Jan 14, 2015 at 04:06:17PM +0000, One Thousand Gnomes wrote: > >> and I think you effectively have the user usage covered here for such >> things. It much like GPIO pins - we can describe them but we can also >> declare they are not visible to the user. > > A missing element in mainline is a kind of VFIO scheme to let user > space access the FPGA registers designated for user space use. > > We have been using these patches since 2.6.16 to allow user space to > access the FPGA register resources via a PCI like /sys/../resource0 > mmapable: > > https://github.com/jgunthorpe/linux/commit/59d5d13ddeffa8980ccc6248ebb5f1678ccb23f4 > https://github.com/jgunthorpe/linux/commit/7c29c4344627be8a3906d64d32db533bc131df86 > https://github.com/jgunthorpe/linux/commit/e41bb4a197368a9d505d66b627aee82f2d2b8895 > > We deliberately split up the FPGA registers and the assign user space > permissions to the resource0 files in a way that makes sense for our > app. Typically there are 10-20 FPGA register regions. User space does > not access register regions that control DMA. > Interesting. >> The swappable case mostly comes out of the /dev node. Once you have the >> dev node it becomes a detail of the OS not the FPGA driver as to who may >> open it, and how it is handed about. It might be an FPU manager daemon it >> might not. > > Right, but the thing that scares me about the swappable case is the > kernel side support.. The FPGA has to connect to the CPU in some way, > it must have some address assigned, etc. Swapping needs to take care > of all those details in some way. > > A fixed bus interface block and dynamic reconfiguration for the > remainder is probably the way to manage that. But, that implies that > even a family of swappable FPGAs will have a DT overlay associated > with it. > > Ideally, I could see wanting to have a file format that combines the > overlay and FPGA bitfile. A loader tool would use the /dev/ interface > to setup both elements. That would be much more robust than the > current scheme I see (eg Xilinx) using where the bitfile and DT bit > live in completely different places and have to be perfectly matched. > A single DT property can be 4 gigs of binary data (cheating a bit, the whole blob can be 4 gigs). You can conceivably include the whole bitfile in the DT blob. Whether this is a perversion or not it’s left to the reader. The bitfile does describe the hardware though. > Jason Regards — Pantelis -- To unsubscribe from this list: send the line "unsubscribe linux-doc" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
